Business School
WORKING PAPER SERIES
Working Paper
2014-071
The Impact of Internal Migration on
Local Labour Markets in Thailand
Eliane El Badaoui
Eric Strobl
Frank Walsh
http://www.ipag.fr/fr/accueil/la-recherche/publications-WP.html
IPAG Business School
184, Boulevard Saint-Germain
75006 Paris
France
IPAG working papers are circulated for discussion and comments only. They have not been
peer-reviewed and may not be reproduced without permission of the authors.
The Impact of Internal Migration on Local Labour Markets in Thailand
Eliane El Badaoui†
Eric Strobl‡
Frank Walsh*
Abstract
We estimate the impact of internal migration on local labour markets in Thailand. Using an
instrumental variable approach based on weather and distance we estimate an exogenous
measure of the net migration inflow into each region. Our results show that instrumenting
for the possible endogeneity of net inward migration is crucial to the analysis. The results
suggest that wages of low skill male workers are highly flexible with substantial adjustments
in hours worked and weekly wages in response to short term changes in labour supply. We
find no effect on high skilled workers.
Key-words: Internal migration, Labour markets, Thailand
JEL-Classifications: O15, J10
†
EconomiX - Université de Paris X, 200 Avenue de la République - 92001 Nanterre Cedex. Tel.: 33 (0) 1 40 97 78
25. Email: [email protected]
‡
Ecole Polytechnique Paris and IPAG Research Lab.
*
University College Dublin.
1
I. Introduction
The 2009 Human development Report [UN (2009)] estimates that at least 740 million people
worldwide are internal migrants, i.e., almost four times the number who have moved
internationally. Despite this, perhaps because movements within borders often go
undocumented, the literature on internal migration and its consequences on local labour
markets is relatively small and not very well understood.1 In this regard, a key difficulty in
measuring the impact of internal migration is the endogeneity of migration. More precisely,
net migration is expected to be correlated with economic conditions in each region making it
difficult to identify the impact of migration on variables such as the wage and employment
level, which also depend on these factors. In this study we use arguably exogenous climatic
shocks to identify the impact of internal migration on the destination labour markets in
Thailand.
The literature on the impact of international migration on labour markets can serve as a
first indication of what effects one might expect from internal migration. For example, well
known studies such as Card (1991) looked at the impact of exogenous regional migration
shocks such as the 1980 Mariel boatlift and found that migration had little impact on native
wages. Critics argued that a possible cause for the absence of any observed effect of
migration on natives is that natives might move to other local labour markets in response to
an influx of migrants, thus masking the impact of migration on wages and employment.
While some studies such as Aydemir and Borjas (2007) use national data to overcome this
problem and find a negative effect of migration on wages, Aydemir and Borjas (2010) note
that “… the national labour market approach may find itself with as many different types of
1
Also see Lucas (1997) or Mendola (2012) for reviews of the literature on internal migration in developing
countries.
2
results as the spatial correlation approach that it conceptually and empirically attempted to
replace2 .” An alternative explanation for the absence of important effects on wages and
employment prospects for natives from an increase in migration is that native and migrant
workers may be imperfect substitutes [Manacorda, Manning and Wadsworth (2006),
Ottaviano and Peri (2012), Peri (2011)]. In particular Manacorda et al. (2006) suggest using
U.K. data that, while migrants and natives are imperfect substitutes, migrants are close
substitutes for other migrants so that an increase in the stock of migrants lowers the wages
of existing migrants but has little impact on natives. Arguably, however, internal migrants
will be closer substitutes for native workers than international migrants so that these effects
are less likely to be as important in our context of Thailand.
The little empirical literature that exists on the effect internal migration on local labour
markets also presents mixed results. For instance, Beals et al. (1967) study the migration
phenomenon in Ghana and show that income differentials drive migration and that regions
of large population are relatively more attractive. Sahota (1968) finds that internal migration
in Brazil is highly responsive to earning differentials and inversely related to distance. More
generally, economic costs and returns dominate the behaviour of migrants. Phan and
Coxhead (2010) analyse inter-provincial migration and inequality during Vietnam’s
transition. Their analysis suggests that the impact of migration on inequality can be either
negative or positive. Boustan et al. (2010) study the effect of internal migration in the 1930s
on the economic welfare of residents in US destination cities and find a small and not
significant impact of in-migration on hourly and weekly earnings while Ham et al. (2011) use
2
Some examples of studies that have examined this question with mixed results are Bonin (2005) who reports
a very weak impact of supply shifts on wages in Germany. Bohn and Sanders find a weak wage effect on the
Canadian labor market. Aydemir and Borjas (2007) use data from Canada and Mexico and find a strong
negative relationship between wages and supply shifts induced by immigration while Mishra (2007) studies the
Mexican labor market and finds a significant positive effect of emigration and wages in Mexico.
3
a distance based measure of migration to measure the effect of internal U.S. migration on
wage growth and find that migration significantly increases wage growth for college
graduates. Kennan and Walker (2011) show that expected income is an important
determinant of interstate migration in the U.S.
In this paper we analyse the effects of inter-provincial migration on wages and
employment in Thailand. Using the methodology developed by Boustan et al. (2010) we
estimate the net inflow rate of migrants into each province thus controlling for the labour
supply responses of natives to changes in migration. We take into consideration the
geographic distance between sending and receiving provinces. In fact, the distance
constitutes an important determinant of the location choice of one migrant and several
studies consider that distance has a strong negative effect on migration.3 Furthermore, we
consider the fact that, in developing economies, weather conditions might induce a spatial
reallocation of the relatively mobile input, labour.4 Thus, local weather conditions can
constitute a convenient and readily available instrument for migration rates (Boustan et al.,
2010).5
We use our framework and data from the Thai Labor Force Survey to identify on the
impact of internal migration on the weekly wage, hourly wage and on weekly hours worked
for males at the destination markets in Thailand. 6 Arguably, Thailand constitutes an ideal
case study for the task at hand. Standards of living, economic and cultural structures, and
growth rates differ widely among provinces. As a consequence, levels and patterns of
3
See, for instance, Sjaastad (1962), Sahota (1968) and Schwartz (1973).
In this study, we consider voluntary climate-induced migration and not climate-forced migration.
5
For instance, Yang & Choi (2007) examine how remittances sent by migrants respond to income shocks
experienced by Philippine households. The authors use rainfall shocks as instrumental variables for income
changes and show that, in households with migrant members, exogenous income declines are partially covered
by foreign remittances. More particularly, households with migrant members enjoy a flat consumption path
compared to households without migrants for whom consumption responds strongly to income shocks.
6
We focus on males because employment rates are much higher than for females and because for the latter
estimation is further complicated by needing to model the labour supply decision.
4
4
internal migration vary within the country. Also, Thailand is one of the earliest Southeast
Asian economies to implement an export-led growth strategy, the consequence of which is
an increase in rural-urban migration, especially to the service sector in Bangkok (Guest,
2003). Despite the 1997 crisis that altered migration patterns for seasonal and short-term
workers, this long-term growth in labour migration has continued.7 The internal migration of
workers is induced by the wage differentials observed among provinces and modifies wages
in receiving provinces. Moreover, migrants benefit from a net income increase that they
share with household members in the form of money and goods they remit home. In this
regard, Yang (2004) analyses the link between migration and cross-province inequality in
Thailand and finds a significant effect of migration on income inequality. More particularly,
she reports that a 1 percent increase in the mean fraction of out-migrants to Bangkok entails
a 0.058 reduction in the average ratio of Bangkok’s income to all other provinces. 8
The results for low skilled males indicate that inward migration has a substantial
negative impact on the weekly wage of natives, but this results from a reduction in weekly
hours worked rather than in the hourly wage. In this regard, one might have expected an
increase in migration to lead to higher hours worked and lower wages in a frictionless labour
market. However, in our theoretical section we use the Mortensen Pissarides (1994) search
model modified to include hours worked by Pissarides (2000 ) to show that an increase in
migration lowers the search costs associated with finding workers in equilibrium and leads to
substitution from hours into workers in line with the empirical results.
7
More particularly, the seasonal migration from the northeast of Thailand, facilitated by wide networks of
friends and relatives, has continued on a large scale (IOM, 2008). This form of migration represents the main
source of remittances for out-migration region.
8
Vanwey (2003) analyses the role of land ownership in rural temporary migration in Thailand.
5
The paper is organized as follows. In the next section we outline the theoretical model,
section three outlines our data set. The fourth section presents the empirical specification
and econometric results. The final section concludes.
II. Theory
Kinoshita (1987) derives the equilibrium properties in the standard competitive model
where firms combine hours per worker with the number of workers in the production
function9. We begin by sketching how an increase in migration might affect equilibrium in
this type of model and continue by outlining how the results might differ when there are
search frictions in the model.
In general the equilibrium hours per worker hourly wage locus is a set of tangencies
where workers who wish to work longer hours match up with like-minded firms. In
equilibrium the supply and demand of each worker type are equal and no worker or firm can
gain from deviating to another point on the locus. Compensating wage differentials are paid
to workers for working a less desirable number of hours10. Firms are assumed to be able to
hire as many workers as they wish at any level of hours (h). If there is only one type of
worker and firm the workers’ indifference curve is the equilibrium hours wage locus and the
firms isoprofit curve over wage hours space.11 Figure 1 illustrates a tangency between the
indifference curve of a representative worker and the isoprofit curve of a firm. We expect
that an increase in Migration (Labour supply) would shift the equilibrium indifference curve
downward and as long as equilibrium is on an upward sloping part of the indifference curve
(so that in equilibrium a firm would pay a higher hourly wage if they wished to induce
9
The following passage draws heavily on the introduction to Strobl and Walsh (2011)
The models of Lewis (1969) and Rosen (1986) are the precursors to this model.
11
The firm chooses the number of workers optimally and we substitute the first order condition on the number
of workers into the profit function to trace out the isoprofit curve
10
6
workers to work longer hours) a lower reservation indifference curve will be tangent to the
isoprofit curve at higher hours per worker and a lower hourly wage. Even in this simplest
version of the competitive model with a representative worker and firm it is theoretically
possible that the equilibrium is on a downward sloping part of the indifference curve where
workers work short hours and would accept a lower hourly wage to work longer hours and
where an increase in labour supply (downward shift in the indifference curve) would lead to
lower hours per worker, although arguably most labour economists might expect there to be
a positive relationship between hours and the hourly wage for most workers in equilibrium.
That is our expectation from the standard model might be that an increase in migration,
where migrants are substitutes for natives, would increase labour supply and lead to lower
wages and longer working hours.
Another prediction of the standard models of the labour market is that an increase in
the fixed costs of hiring workers can cause firms to substitute from workers into longer
hours per worker [See Hamermesh (1993) Chapter 7 for example]. Most of the modern
labour labour literature models labour market flows in a framework where there are search
costs associated with firms and workers meeting. In the model presented here we use one of
the most common search models: the Mortensen Pissarides (1994) search model with hours
worked included, as outlined in Pissarides (2000) and analyse the impact of migration on
hours and wages. The intuition is that migration may lower the time and cost of contacting
workers and lead to substitution from hours into workers so that hours per worker falls in
contrast to the expected prediction from the competitive model outlined above.
Below we present the search model outlined in Pissarides (2000) section 7.3. We
incorporate a fixed benefit payment to unemployed workers into the model and analyse the
impact of an increase in migration (an increase in the labour force) on hours and the number
7
of workers in equilibrium. We show that an increase in migration is associated with a
decrease in labour market tightness so that it takes less time for firms to find workers. This
increases the employment rate in a stationary equilibrium and reduces the number of hours
per worker.
There is a matching function which is homogeneous of degree one so that the number
of matches taking place at each unit of time is:
mL  m(uL, vL)
(1)
The labour force is L, the unemployment rate is u and the number of vacancies as a fraction
of the labour force is v. We define the ratio of unemployment to vacancies (labour market
tightness as):  
v
. The rate at which vacant jobs are filled can be rewritten:
u
u
q( )  m( ,1)
v
The mean duration of a vacancy is:
(2)
1
and the percent change in q( ) from a percent
q ( )
change in  is 1    0 . Similarly the rate at which unemployed workers find jobs is:
 q( ) 
m(uL, vL)
uL
The mean duration of unemployment is:
(3)
1
and the elasticity is 1   . Idiosyncratic
 q( )
shocks destroy jobs at a rate  . With a fixed labour force the number of workers who enter
unemployment during a short interval  t is:
 (1  u) L t
The mean number of workers who leave unemployment at each point in time is:
8
(4)
mL t  u q( ) L t
(5)
The unemployment rate in the steady state is:
u

   q( )
(6)
It also follows that if the labour force is L then aggregate employment is:
N  (1  u ) L 
 q( ) L
   q( )
(7)
Equation (6) is the Beveridge curve which can be used to graph the relationship between the
vacancy rate and unemployment rate in long run equilibrium. This is independent of the size
of the labour force. The workers instantaneous utility depends on current income and
current hours of work. Instantaneous utility during employment is:
wE  wh (1  h)
'(.)>0,''(.)  0, 0  h  1
12
(8)
Where the hourly wage is w, the length of the day is normalised to unity and the utility
function is linear in wage income and non-linear in the disutility of work. The flow utility
from unemployment and employment are:
rU  z   q( )(W  U )
(9)
The model differs from the model in Pissarides (2000) in that we include a fixed
payment z to unemployed workers. Pissarides (2000) discusses the possibility of making this
payment proportional to the wage and we might reasonably argue that benefits are a
function of earnings in many countries. While this is reasonable in terms of analysing long
run equilibrium, we suggest here that if we wish to analyse the short run effects of migration
 '(.) is the marginal utility of leisure which is positive. In the analysis we will differentiate with
 (1  h)
respect to work hours where
  '(1  h)
12
We note that
h
9
we do not expect benefits to respond instantaneously to any change in wage associated with
short run migration flows. This is important for the results in that if benefits are fixed the
wage will not be fully proportional to productivity. Because wages are not proportional to
productivity there will be a relationship between the equilibrium level of productivity and
labour market tightness. This means that if an increase in migration affects employment and
the equilibrium level of productivity, this will also lead to an effect on labour market
tightness and search costs. We will see that this leads to a change in hours per worker.
rW  wh (1  h)   (U  W )
(10)
Wages and hours are chosen by Nash bargaining where the flow value of a vacancy is:
rV   p[ Nh]c  q( )( J  V )
(11)
The cost of maintaining a vacancy per unit of time is pc while there is a probability of q( )
that the vacancy will be filled. We assume that productivity is the constant p[Nh] but that
the productivity of additional matches is diminishing at the aggregate level, so that the
change in productivity from a change in aggregate hours worked NH is negative. The flow
value of a filled job is:
rJ  p[ Nh]h  hw   J
(12)
Hours and the wage are chosen to maximise:
(W  U )  ( J j  V )1 
0< <1
(13)
(
wh (1  h j )  rU
(r   )
) (
[ p( N , h)  w]h
 V )1 
(r   )
We note that when the worker and firm engage in bargaining at the level of the individual
worker/job match, the value of the outside option is exogenously given to the worker as is
10
the value of a vacancy to the firm. The first order conditions for the wage and hours of work
satisfy:
 (J  V )
 (W  U )
( J  V )
 (1   )(W  U )
w
w
(14)
  ( J  V ) (1  h)  (1   )(W  U )  0
And:
 ( J  V )w (1  h)[1 
 '(1  h)h
]  (1   )(W  U )[ p( N , h)  w]
 (1  h)
(15)
Firms continue to create vacancies until the marginal product of an additional vacancy is
zero so that:
J
pc
q( )
(16)
From (9) and (10) we calculate:
W U 
wh (1  h)  z
r     q( )
(17)
r
pc
q( )
(18)
Substituting (16) in (12) we get:
[ p( Nh)  w]h 
Hence by using V=0, the value of J from (16), the value of W-U from (17) and using
[ p( Nh)  w]hq( )
 r   from (18), we get the wage equation from the first order condition
pc
on the wage [equation (14)] as:
w   p( N , h)[1 
11
c
h
]  (1   )
z
h (1  h)
(19)
Using the value for  ( J j  V ) (1  h) from (14) in (15) we see that h is chosen such that:
w  p( Nh)
 (.)
 '(.)h
(20)
The job creation equation is (18) and can be rewritten in terms of the wage as:
w  p( N , h)[
q( )h  (r   )c
]
q( )h
(21)
We continue by assuming the Cobb-Douglas utility function for hours:
 (1  h)  (1  h)
(22)
Equation (20) can be rewritten in this case as:
w  p( E )
 (.)
 (.)
p( Nh) 1  h
 p( Nh)

 '(.)h
 '(.)h

h
(23)
Equating equations (23) and (21) we can solve for hours as:
h
1
 ( r   )c
1
(q( )   (r   )c


[
]
(1   ) (1   ) q( )
(1   )
q( )
(24)
We note that since q '( )  0 :
h
 (r   )cq '( )

0

(1   )
q( )2
(25)
Hours per worker increases with labour market tightness. The implication of this is that if an
increase in migration leads to a fall in labour market tightness, hours per worker will fall. We
also note that:
1 h 
 q( )  (r   )c
[
]
1
q( )
(26)
Using equation (7) for the aggregate number of workers and equation (24) for hours per
worker we get:
12
Nh 
L
q( )  (r   )c
[

]
   q( ) (1   )
(1   )
(27)
We note from (7) that:
q '( )
]
N
q( )
  q( )
0

[   q( )]2
[1  
Since:
(28)
d ( Nh) N
h

h
N it follows that:
d


dE d ( Nh)

0
d
d
(29)
Equating (21) and (19) yields the following solution for productivity per worker:
p ( E )  p[ N ( )h( )] 
(1   ) z
[(r     q ( )]c
[1  h( )] {(1   ) h 
}
q ( )
(30)

(1   ) z
(1   ) z


1



(1


)(
r


)
c
[1  h( )] {
  
} [1  h( )] X ( )
1
q( )
To get the relationship between migration and labour market tightness we totally
differentiate (30) over the labour force and labour market tightness:
p( E ) E
p( E ) E { (1  h) 1 X ( )  X '( )[1  h( )] }
dL  {

(1   ) z}d
E L
E 
[1  h( )]2 X ( ) 2
Rearranging equation (30) we can write the derivative of labour market tightness from an
increase in migration:
d

dL
p( E ) E
E L
p( E ) E { (1  h) 1 X ( )  X '( )[1  h( )] }
{

(1   ) z}
E 
[1  h( )]2 X ( ) 2
13
(31)
We note that since
p( E )
N
 q( )
 0 by assumption and since

 0 from (7) then
E
L    q( )
the numerator of equation (30) is negative. The first term in the denominator: 
positive since 
p( E ) E
is
E 
p( E )
E
 0 by assumption while
 0 from (7). Since hours are less than
E

one  (1  h) 1  0 while from (29) positive productivity implies X ( )  0 . Differentiating
X ( ) from (29) we get:
X ( )
 (1   )(r   )c
 1  q '( )
0

q( )2
(32)
It follows that the term  X '( )[1  h( )] in the denominator of (30) is also positive. Finally
since [1  h( )]2 X ( )2  0 and (1   ) z  0 we have shown that the numerator of (30) is
negative while all the various terms in the denominator are positive so that:
d
0
dL
(33)
The derivative of hours per worker with respect to a change in migration is:
dh d h

dL dL 
Since
(34)
h
 0 from (25) and given equation (32) an increase in migration will lower labour

market tightness and cause firms to reduce hours per workers. We can also derive the
change in employment from a change in migration which is positive as we expect:
q '( )
]
dN N N d
 q( )
q( ) d



  q( )
0
dL L  dL    q( )
[   q( )]2 dL
[1  
14
(35)
The change in the employment rate from a change in migration can be derived by
differentiating equation (6):
d (1  u )

dL
[

q '( )
]
q( )[1 
]
   q( ) d
q( ) d

0

dL
[   q( ]2 dL
(36)
Equations (35) and (36) demonstrate that employment increases for two reasons after an
increase in migration. The increased number of workers increases employment at a given
employment rate, but the reduction in labour market tightness increases the employment
rate also. An increase in migration leads to lower hours and a higher employment rate.
III. Data
We use data on males from the Thai Labour Force Survey between 1991 and 2000. The
survey is conducted several times a year and with increasing frequency in recent years. We
have data for the February and August surveys for each year. The survey is a large crosssection; for example, the February 2000 survey has 164,636 observations. In sum, the
complete data is constituted of twenty waves with a total of 2,951,839 observations. August
is the peak of the agricultural season and labour markets are much more buoyant with
significant internal migration as workers return to rural areas for harvesting. February clearly
represents the off-peak season in Thailand.
The survey contains a wide variety of questions on location, employment status and job
characteristics, income as well as demographic characteristics. Data on earnings asks
workers if they are paid hourly, daily, weekly, or monthly and what the rate of pay is for the
relevant category. Summary statistics for the total sample shows that 91% of workers are
15
paid either daily or monthly, where most waged workers at the low skill end of the labour
market are paid daily.
There are seventy two provinces in Thailand as shown in Figure 1. In addition to
providing the name of the province where they live, individuals answer the following
question: “How long have you been living regularly in this village/ municipality?”
Respondents can answer from less than a year, one year, two years, up until nine or more
than nine years. We exclude people who move into the same province and calculate the
number of recent arrivals as those who answer less than or equal to one year. This number
corresponds to 124,185 individuals and represents 52.4% of total movers to new
provinces.13 We use this subsample of movers to compute the inflow and outflow rates.
Then we define the province of origin and the destination province of all movers as people
are asked “Which is the previous province of your residence before moving here?”. The
survey then asks for the reason of migration. Among recently moved people, some 35.71%
were looking for a job or occupation. However, 7.62% of respondents migrate for further
study, 22.75% follow their family and 28.53% report coming back to their former residence.
Only 0.22% of migrants state moving from one province to another in order to be nursed.
Concerning the province of destination, Bangkok counts the largest proportion of arrivals
with 7.2% of total recent migrants.14
We construct a sample of non-migrants residing in different areas – municipal, nonmunicipal, sanitation district – in the 72 Thai provinces. We consider as incumbents all
recent migrants who moved within the same province and we reduce our sample to the
working age population. Then we reduce the sample to men who were not attending school
13
Note that the category of movers within provinces represents 29% of all movers and that 49.4% of the
sample of movers from this category moved one year ago at most.
14
The second best destination province is Udon Thani with 4.26% of total recent migrants.
16
at the moment of the survey and who work 95 hours or less a week in the private sector.
Moreover, we drop observations of workers who are in unpaid jobs. The size of the sample
used in all the wage and hours regressions below is 194,410 and recent migrants represent
10.11% of the sample. Table A1 provides summary statistics separately for the subsamples of
incumbents and recent migrants for whom all the variables are non-missing.
III. Econometric Analysis
A. Construction of Instruments
In order to construct instruments for migration we follow the methodology proposed by
Boustan et al. (2010), which consists of predicting the total outflow (inflow) from a region
induced by weather shocks, and then decomposing this outflow (inflow) into destination
regions by estimating the role of geographic distances in determining inter regional flows.
We then use both weather and distance to construct the predicted inflow (outflow). More
specifically, for the case of migration inflow this first involves regressing total outflow rates
of each region on a set of climate determinants:
(37)
where
is the outflow rate from source region over time period
is a vector of climate specific indicators, and
is an error term. Using the estimated
coefficients from (1) the predicted flow of migrants leaving each region ,
just equal to the predicted outflow rate,
to ,
, times the population at
, is then
.
(38)
17
One then separately for each sending area i regresses the actual set of destination
specific outflow rates to each destination region j on their relative distances and it’s squared
and cubic value15:
(39)
The instrument for in-migration to region j ,
number of migrants over all areas (
, is then just the sum of the predicted
),
expected to settle in region j:
(40)
One can then in a similar manner construct predicted outflow from area j by predicting
the in-migration rates to each receiving area i using climatic determinants, using these rates
to predict the number of inflowing migrants into i, and then constructed predicted outflow
migrants by multiplying this figure by the distance and its non-linear terms estimated
inflowing rates between regions i and j (
).
In order to estimate (1), as well as its analogous specifications for the in-migration, we
use for the vector Z a number of measures that capture weather conditions in a region. In
order to identify periods of extreme wetness and dryness in regions we first calculated the
local standardized precipitation index (SPI), which has been argued to be particularly good at
capturing the cumulative effect of high and low patterns of rainfall over time in a chosen
locality, from the mean monthly precipitation values within our regions as calculated from
the IPCC data set.16 Following McKee et al. (1993) we then define a monthly extremely dry
(wet) event as starting when the SPI reaches an intensity of -2.0 (2.0) or less (more) and as
15
One should note that Boustan et al. (2010) regress these rates only on distance and its squared value. For the
case of Thailand we found that including its cubic value substantially increased the specifications fit.
16
The calculation of the SPI is based on modeling the probability distribution of precipitation as derived from long term
records by fitting these to a gamma distribution via maximum likelihood. An important component in this regard is the
chosen time scale. Since we are interested in cropland productivity and soil moisture conditions are known to respond to
precipitation anomalies over a relatively short time period, we use a 12 month scale.
See http://www.drought.unl.edu/whatis/indices.htm.
18
ending once the index become positive (negative) again. For each time period we then
calculate the number of months of extreme dryness (wetness). The corresponding
constructed variables are DRY and WET, respectively. To capture the effect of temperature,
in particular with respect to its importance for agriculture, we construct a measure of
reference evapotranspiration (ET) to represent the evaporative demand of the air within a
basin. Following Hargreaves and Samani (1985), evapotranspiration is calculated as:
ET = 0.0023 (Tavg + 17.8) (Tmax - Tmin) 0.5 Ra
(41)
where Tavg, Tmax and Tmin are mean, maximum and minimum temperature, respectively and
Ra is the extraterrestrial radiation calculated following Allen et al. (1998). Since the effects of
rainfall shortages and abundance on local agricultural are likely to some extent to depend on
the local evaporative demand, we also allow for interactions between ET and WET and DRY.
To construct all these climatic factors at the regional level we resort to information from the
Inter-Governmental Panel on Climate Change (IPCC) climatic data set, which provides
monthly precipitation and temperature measures across the globe at the 0.5 degree level
over the entire 20th century. We use these to calculate out time varying averages within
provinces.
The results of estimating (37) for the annual regional out- and in-migration rates,
controlling for regional specific fixed effects and region common time specific factors are
given in Table 1a. Moreover, we calculate Driscoll and Kray (1998) standard errors corrected
for spatial correlation throughout. As can be seen, for both of inflow and outflow rates, the
set of climatic variables are almost all significant, producing highly significant F-tests of joint
significance. Examining the individual factors, one finds that for the precipitation related
factors the signs meet a priori expectations. More specifically, one finds that extremely dry
as well as extremely wet weather, indicative of drought and flood like conditions,
19
respectively, act to increase overall outflow from regions. Moreover, the negative impact of
rainfall shortage is further exacerbated by a high evapotranspirative demand of the air.
Somewhat surprisingly, the direct effect of evapotranspiration is to reduce outflow from a
region, although in absolute terms this impact is small. For the inflow rate, one finds that
extremely wet periods tend to reduce the inflow rate, while droughts have no significant
effect. Furthermore, a high evapotranspirative demand of the air tends to reduce the effect
of the latter. Surprisingly one finds that this demand on its own acts to increase person
flowing to the region, although again not substantially so. To construct the predicted inward
and outward migration rates by sub-group we proceeded in similar manner as for the overall
sample, except restricting construction via (37) through (40) to the sub-sample in question.
We report the estimation for (37) for the outflow and inflow rates in Tables 1b and 1c,
respectively. As can be seen, for the outflow rate all climatic variables are significant, where
the signs are in congruence with the overall sample. Unsurprisingly the joint F-tests attests
to their power as predictive factors. For the inflow rates, the majority of coefficients are
significant and similar to those from the overall sample. Similarly, the F-test statistics provide
evidence of their predictive power.
In terms of estimating (39), since this involves estimating different specifications for
each region, we only provide a brief outline of the results. One may want to first note that
since our distance measures do not vary over time, our estimated specification in (39) does
not control for region specific effects, but does include a set of time dummies to control for
common region time specific factors determining the migration flows. We used Driscoll and
Kray (1998) standard errors corrected for spatial correlation as we did for (37). For each
region specific regression, we, after estimating the parameters on distance conducted an Ftest of the null hypothesis that these were jointly zero. In the case of out-migration rates for
20
only 4 regions, while in the case of in-migration rates for only 6 could the null hypothesis not
be rejected. As with the overall sample the F-test of the distance variables suggested strong
predictive power in almost all cases for the estimation of (39) for subgroups.
In Table 2 we depict the results from the first stage regression where we use predicted
migration rates constructed as outlined above to predict actual net migration rates. Table 2
shows the results for men by skill. As can be seen, and is indicated by the F-Test on the
instruments, the predicted inflow rate variables significantly predict an increase in actual net
migration, whereas predicted out-migration rate acts to decrease net migration. Using
bootstrapped standard errors and the corresponding Wald tests show similar results
although standard errors are somewhat larger.
B. The Effect of Net Migration on the Local Labour Market
We next examine the impact of net inward migration on the weekly wage, weekly hours
and hourly wage for males of working age (15-64 controlling for individual characteristics. In
particular when we look at the impact of migration on the wage we control for age and age
squared, marital status (four dummies indicating single, married, widowed or divorced
status), a set of twelve educational indicator dummies, ten occupation dummies and ten
industry dummies, seven firm size dummies, dummies indicating whether the worker is paid
hourly, daily, weekly or monthly and a dummy indicating whether the worker lives in a
municipal area. We also include average age and the fraction of workers with no education
at the province level as well as province and time specific effects. The results on the
estimated coefficient on the net inward migration rate for weekly wage and weekly hours
and the hourly wage are reported in Table 3. So for example the coefficient of -.088 on
weekly wages in the OLS results in Table 3 indicates that an increase in the net migration
21
rate of 10% (which means the working age population increases by 10% due to net
migration) is associated with a decrease in male wages by 8.8%. A clear worry here is that,
as noted earlier, we might expect migration inflows to depend on local economic conditions
which also affect the wage level or level of hours worked. The results from the instrumental
variables regression confirm that this worry is legitimate, showing a weekly wage decrease
of 31.5% for males when migration increases population by 10%. We also note that while
OLS results indicate a small insignificant coefficient on hours worked the instrumental
variables results indicate that a 10% increase in population by migrants lowers weekly hours
by 11.753, i.e., a decrease of about 22% based on the summary statistics.
Table 3 provides results by high and low skill group. This makes a substantial difference
to the OLS results. Skill specific migration is negatively associated with weekly wages for
high skilled workers with a 8.6% fall in wages for a 10% increase in population. Low skill
weekly wages fall by 5.3% for a similar increase in migration17. Coefficients on weekly hours
are small and statistically insignificant for both skill groups.
Once again instrumenting
makes a substantial difference to the results. For the instrumental variables analysis weekly
wages of skilled workers are unaffected by migration while unskilled weekly wages fall by
53.4% in response to a 10% increase in the population. The coefficient on hourly wage is
statistically insignificant while hours worked fall by 18.839 if population increases by 10%.
This represents a decrease of 36% in the average hours of low skilled incumbents.
17
We explored the possibility that low/high skill worker’s wages and hours might be affected by the migration
rate of the other group: that is that to explore whether these groups were complements/substitutes for one
another. Unfortunately migration rates for are highly correlated across groups (even when instrumented)
making it difficult to identify a separate effect.
22
IV. Conclusion
In this paper we have examined the impact of net inward migration on local labour
markets in Thailand. To this end we have constructed a data set of regional migration flows
and individual labour market outcomes for the period 1991 to 2000 using the Thai Labour
Force Survey. Our results show that instrumenting for the possible endogeneity of net
inward migration is crucial to the analysis. The results suggest that wages of low skill male
workers are highly flexible with substantial adjustments in hours worked and weekly wages
in response to short term changes in labour supply. We find no effect on high skilled
workers. It may be that wages are slower to adjust for skilled workers due to implicit
contracts, firm specific capital or other institutional features that limit firms’ ability or
willingness to adjust wages in response to possibly temporary shock.18
18
See Beaudry and Dinardo (1991) for an example and some evidence for an implicit contracts model while Hall
(2005) shows that the local monopoly rents in a search matching model mean that wages can be sticky without
violating rationality.
23
Table 1a: The Effect of Weather on Migrant Flows
DRY
WET
EVAPO
EVAPO*WET
EVAPO*DRY
Observations
Number of groups
F-test
OUTRATE
INRATE
0.00266*
(0.00130)
0.00352*
(0.00135)
-0.000222*
(8.90e-05)
3.04e-05
(1.55e-05)
3.36e-05*
(1.56e-05)
-2.45e-05
(0.000964)
-0.00358**
(0.000701)
0.000710**
(0.000192)
-5.95e-05**
(1.16e-05)
-1.34e-05
(6.74e-06)
1,440
72
4.963
1,440
72
8.616
Notes: (i) Driscoll and Kray (1998) standard errors corrected for
spatial and autocorrelation in parentheses; (ii) ** and * are 1 and 5
per cent significance levels; (iii) year and binannual dummies included
but not reported; (iv) F-test is test of joint significance of the climatic
variables.
24
Table 1b: The Effect of Weather on Outflow Rates
(1)
(2)
(3)
0.00142*
(0.000625)
0.00235**
(0.000709)
-0.000227**
(7.61e-05)
2.75e-05**
(7.25e-06)
1.88e-05*
(7.68e-06)
0.00166*
(0.000648)
0.00231**
(0.000692)
-0.000262**
(8.85e-05)
2.54e-05**
(7.64e-06)
2.14e-05*
(9.34e-06)
0.00120
(0.000618)
0.00238**
(0.000736)
-0.000197**
(6.81e-05)
2.95e-05**
(7.35e-06)
1.65e-05*
(6.40e-06)
Sample
Male
Observations
Provinces
F-test
1,440
72
9.028
Low-Skilled
Men
1,440
72
8.172
High-Skilled
Men
1,440
72
9.926
DRY
WET
EVAPO
EVAPO*WET
EVAPO*DRY
Notes: (i) Driscoll and Kray (1998) standard errors corrected for spatial
and autocorrelation in parentheses; (ii) ** and * are 1 and 5 per cent
significance levels; (iii) year and binannual dummies included but not
reported; (iv) F-test is test of joint significance of the climatic variables.
25
Table 1c: The Effect of Weather on Inflow Rates
(1)
(2)
(3)
-0.000417**
(0.000153)
-0.000891*
(0.000366)
0.000636**
(0.000182)
-8.68e-06
(6.42e-06)
-9.25e-06
(1.10e-05)
-0.000562**
(0.000129)
-0.000797*
(0.000325)
0.000613**
(0.000163)
-5.18e-06
(5.34e-06)
-9.87e-06
(7.63e-06)
-0.000375**
(0.000129)
-0.000764**
(0.000266)
0.000439**
(0.000160)
-9.63e-06
(4.97e-06)
-9.74e-06
(6.74e-06)
Sample
Male
Observations
Provinces
F-test
1,440
72
18.15
Low-Skilled
Men
1,440
72
31.16
High-Skilled
Men
1,440
72
8.963
DRY
WET
EVAPO
EVAPO*WET
EVAPO*DRY
Notes: (i) Driscoll and Kray (1998) standard errors corrected for spatial and
autocorrelation in parentheses; (ii) ** and * are 1 and 5 per cent significance
levels; (iii) year and binannual dummies included but not reported; (iv) Ftest is test of joint significance of the climatic variables.
26
Table 2: Relationship between Predicted and Actual Migration for Men by Skill
WLS Regression
Predicted in-migration rate
Predicted out-migration rate
F-Statistic
Bootstrapped Procedure
Predicted in-migration rate
Predicted out-migration rate
Wald’s Statistic
Total Sample
Low-Skilled
Subsample
High-Skilled
Subsample
5.332**
(2.592)
-9.537***
(2.302)
15.1
3.007**
(1.300)
-4.101***
(1.128)
16.1
3.262**
(1.387)
-5.539***
(1.219)
12.4
5.576*
(3.088)
-9.591***
(2.756)
2057
3.103**
(1.424)
-4.168***
(1.243)
1704
3.491**
(1.549)
-5.490***
(1.398)
3535
Notes: (i) Regressions are estimated using individual data from the Thai LFS from 1991
to 2000; (ii) Standard errors are clustered by provinces and waves; (iii) Dummies for
salaries period of payment (hourly, daily, weekly and monthly) are introduced. (iv) For
wage regressions, we introduce the number of weekly working hours and its squared
term; (v) We include both the high-skilled and low-skilled net instrumented migration
rate for each of the six subsamples shown above; (vi) ***, ** and * denote significance
at the level of 1%, 5% and 10% significance levels, respectively.
27
Table 3: Effect of Net Migration on Weekly Wages, Hourly Wages and Work Time
- Sample of Incumbent Working Men in the Private Sector Total Sample
Low-Skilled
Subsample
High-Skilled
Subsample
-0.088***
(0.030)
-0.105**
(0.042)
0.610
(2.086)
-0.053*
(0.031)
-0.073*
(0.042)
0.697
(2.447)
-0.086**
(0.034)
-0.124**
(0.057)
1.445
(2.260)
-0.315***
(0.118)
-0.171
(0.129)
-11.753**
(5.449)
-0.534**
(0.222)
-0.334
(0.210)
-18.839**
(8.962)
-0.168
(0.123)
-0.153
(0.144)
-0.817
(4.057)
194,410
130,049
64,361
Ordinary Least Squares
ln(weekly wage)
ln(hourly wage)
Hours worked
Instrumental Variable
ln(weekly wage)
ln(hourly wage)
Hours worked
Observations
Notes: (i) Regressions are estimated using individual data from the Thai LFS from
1991 to 2000; (ii) Standard errors are clustered by provinces and waves; (iii)
Dummies for salaries period of payment (hourly, daily, weekly and monthly) are
introduced. (iv) For weekly wage regressions, we introduce the number of weekly
working hours and its squared term; (v) With the instrumental variable technique,
we include the net instrumented migration rate specific for each subsample; (vi)
***, ** and * denote significance at 1%, 5% and 10% significance levels,
respectively.
Table 4: Mean and Standard Deviation of Migration Rate
- Subsamples of High-skilled/Low-skilled Men Mean
Men
Low Skilled Men
High Skilled Men
0.00736
0.00539
0.00629
28
Standard Deviation
0.04684
0.03963
0.04403
Figure 1: Indifference and Isoprofit curves
29
Figure 2: Thai Provinces
30
References
Allen, R.G., L.S. Pereira, D. Raes and M. Smith, 1998. “Crop evapotranspiration: Guidelines
for computing crop requirements”, Irrigation and Drainage Paper, No. 56, FAO, Rome, Italy.
Aydemir, A. and G. Borjas, 2010. “Attenuation Bias in Measuring the Wage Impact of
Immigration”, Journal of Labor Economics, Vol. 29, No. 1, pp. 69-112.
Aydemir, A. and G. Borjas, 2007. “Cross-Country Variation in the Impact of International
Migration: Canada, Mexico, and the United States”, Journal of the European Economic
Association, Vol. 5, No. 4, pp. 663-708.
Beals, R., M. Levy, and L. Moses, 1967. “Rationality and Migration in Ghana”, Review of
Economics and Statistics, Vol. 49, No. 4, pp. 480-486.
Beaudry, P., and J. Dinardo, 1991. “The Effect of Implicit Contracts on the Movement of
Wages Over the Business Cycle: Evidence From Micro Data”, Journal of Political Economy,
Vol. 99, No. 4, pp. 665-688.
Bohn, S. and S. Sanders, 2007. “Refining the Estimation of Immigration's Labor Market
Effects”, Mimeo.
Bonin, H., 2005. “Wage and employment effects of immigration to Germany: evidence from
a skill group approach”, IZA Working Papers, No. 1875, IZA, Bonn.
Boustan, L. P., V. Fishback, and S. Kantor, 2010. “The effect of internal migration on local
labor markets: American cities during the great depression”, Journal of Development
Economics, Vol. 28, No. 4, pp. 719-746.
Card, D., 1990. “The Impact of the Mariel Boatlift on the Miami Labor Market”, Industrial and
Labor Relations Review, Vol. 43, No. 2, pp. 245-257.
Driscoll, J.C. and A.C. Kraay, 1998. “Consistent Covariance Matrix Estimation with Spatially
Dependent Panel Data”, Review of Economics and Statistics, Vol. 80, No. 4, pp. 549-560.
Fishback, P., W. Horrace, and S. Kantor, 2006. “The impact of New Deal expenditures on
mobility during the Great Depression”, Explorations in Economic History, Vol. 43, No. 2, pp.
179-222.
Funkhouser, E., 2009. “The Choice of Migration Destination: A Longitudinal Approach using
Pre-Migration Outcomes”, Review of Development Economics, Vol. 13, No. 4, pp. 626-640.
Guest, P. 2003. “Bridging the Gap: Internal Migration in Asia”, Mimeo.
Hall, R.E., 2005. “Employment Fluctuations with Equilibrium Wage Stickiness”, American
Economic Review, Vol. 95, No. 1, pp. 50-65.
31
Ham, J.C., L. Xianghong and B.R. Patricia, 2011. “Matching and semi-parametric IV
estimation, a distance-based measure of migration, and the wages of young men”, Journal
of Econometrics, Vol. 161, No. 2, pp. 208-227.
Hamermesh, 1993. Labor Demand, Princeton University Press.
Hargreaves, G.H. and Z.A. Samani, 1985. “Reference crop evapotranspiration from
temperature”, Applied Engineering in Agriculture, Vol. 1, No. 2, pp. 96-99.
Kennan, J. and J. R. Walker, 2011. “The Effect of Expected Income on Individual Migration”,
Econometrica, Vol. 79, No. 1, pp. 211-251.
Kinoshita, T., 1987. “Working Hours and Hedonic Wages in the Market Equilibrium”, Journal
of Political Economy, Vol. 95, No. 61, pp. 1262-1277.
Lucas, R., 1997. “Internal Migration in developing Countries”, Handbook of Population and
Family Economics. Edited by M.R. Rosenzweig and 0. Stark.
Mendola, M., 2012. “Rural Out-Migration and Economic development at Origin: A Review of
the Evidence”, Journal of International Development, Vol. 24, No. 1, pp. 102-122.
Manacorda, M., A. Manning, and J. Wadsworth, 2006. “The impact of immigration on the
structure of male wages: theory and evidence from Britain”, IZA Discussion Papers, No.
2352.
McKee, T, N. Doesken and J. Kleist, 1993. “The relationship of drought frequency and
duration to time scales”. Preprints, 8th Conference on Applied Climatology, pp. 179-184.
January 17-22, Anaheim, California.
Mishra, P., 2007. “Emigration and wages in source countries: Evidence from Mexico”, Journal
of Development Economics, Vol. 82, pp. 180-199.
Ottaviano, G., and G. Peri, 2012. “Rethinking the Effects of Immigration on Wages”, Journal
of the European Economic Association, Vol. 10, No. 1, pp. 152-197.
Peri, G., 2011. “Rethinking the area approach: Immigrants and the labor market in
California”, Journal of International Economics, Vol. 84, No. 1, pp. 1-14.
Phan, D. and I. Coxhead, 2010. “Inter-provincial migration and inequality during Vietnam's
transition”, Journal of Development Economics, Vol. 91, No. 1, pp. 100-112.
Pissarides, C., 2000. Equilibrium Unemployment Theory, Second Ed. MIT Press.
Sahota, G., 1968. “An Economic Analysis of Internal Migration in Brazil”, Journal of Political
Economy, Vol. 76, No. 2, pp. 218-45.
32
Schwartz, A., 1973. “Interpreting the effect of distance on migration”, Journal of Political
Economy, Vol. 81, No. 5, pp. 1153-1169.
Sjaastad, L.A., 1962. “The Costs and Returns of Human Migration”, Journal of Political
Economy, Vol. 70, No. 5, pp. 80-93.
Strobl, E. and F. Walsh, 2011. “The ambiguous effect of minimum wages on hours”, Labour
Economics, Vol. 18, pp. 218–228.
UN Human 2009 Development Report: “Overcoming barriers: Human mobility and
development”.
Vanwey, L.K., 2003. “Ownership as a Determinant of Temporary Migration in Nang Rong,
Thailand”, European Journal of Population, Vol. 19, No. 2, pp. 121-145.
Yang, L., 2004. “Unequal provinces but equal families? An analysis of inequality and
migration in Thailand”, Mimeo.
Yang, D. and H. Choi, 2007. “Are Remittances Insurance? Evidence from Rainfall Shocks in
the Philippines”, World Bank Economic Review, Vol. 21, No. 2, pp. 219-248.
33
Table A.1: Summary Statistics
Incumbents
Variable
Recent Migrants
Mean
SD
Mean
SD
1151.70
51.47
1215.86
11.16
1021.22
52.63
902.88
10.97
Salary Pay Period
Hourly
Daily
Weekly
Monthly
0.00
0.51
0.01
0.48
0.05
0.50
0.09
0.50
0.00
0.53
0.01
0.46
0.05
0.50
0.09
0.50
Age
Single
Married
Widow
Divorced
Separated
33.58
0.31
0.66
0.01
0.01
0.01
10.97
0.46
0.48
0.10
0.09
0.11
27.75
0.47
0.51
0.00
0.00
0.01
8.83
0.50
0.50
0.06
0.07
0.11
Education level
None
Less than Pratom 4
Lower elementary
Elementary
Lower secondary
Upper secondary
Lower vocational
Upper and higher vocational
University academic
University technical vocational
Teacher training
Short Course vocational
Other
0.03
0.02
0.40
0.22
0.14
0.05
0.00
0.05
0.04
0.04
0.01
0.00
0.00
0.16
0.15
0.49
0.42
0.35
0.22
0.00
0.22
0.20
0.19
0.09
0.01
0.02
0.01
0.01
0.26
0.35
0.16
0.08
0.00
0.04
0.03
0.04
0.01
0.00
0.00
0.12
0.12
0.44
0.48
0.37
0.26
0.01
0.20
0.17
0.21
0.08
0.01
0.02
Firm size
1-4 employees
5-9 employees
10-19 employees
20-49 employees
50-99 employees
100-199 employees
200+ employees
0.24
0.24
0.16
0.11
0.06
0.16
0.03
0.43
0.42
0.36
0.32
0.24
0.37
0.18
0.18
0.18
0.15
0.14
0.08
0.22
0.05
0.39
0.39
0.35
0.34
0.28
0.41
0.22
Occupation
Professional Occupation
Administrative Officer
Financial and Accounting Clerks
Wholesale/Retail Trader/Owner
Farmer/Fisherman/Hunter
Miner/Quarry Worker
Transportation
Cotton Spinner/Weaver/Knitter
Type Cutter/Printer/Bookbinder
Services/Sports
0.05
0.02
0.08
0.05
0.13
0.00
0.11
0.36
0.15
0.05
0.22
0.13
0.27
0.22
0.33
0.05
0.31
0.48
0.35
0.23
0.05
0.01
0.07
0.06
0.09
0.00
0.07
0.41
0.16
0.08
0.21
0.10
0.25
0.23
0.28
0.04
0.26
0.49
0.37
0.27
Weekly Wage
Weekly Working Hours
34
Industry
Agriculture, Forestry and Fishing
Mining
Manufacturing
Rubber
Construction
Sanitary Services
Commerce
Transportation
Services
Others
0.13
0.01
0.12
0.17
0.25
0.00
0.17
0.05
0.10
0.00
0.33
0.09
0.32
0.38
0.43
0.03
0.37
0.22
0.31
0.02
0.09
0.01
0.14
0.22
0.25
0.00
0.14
0.03
0.12
0.00
0.29
0.08
0.35
0.42
0.42
0.04
0.35
0.18
0.32
0.02
Union dummy
Municipal area
0.05
0.41
0.21
0.49
0.04
0.45
0.21
0.50
Region
Bangkok
Central region
North region
Northeast region
South region
0.12
0.36
0.19
0.18
0.15
0.33
0.48
0.39
0.38
0.35
0.12
0.41
0.16
0.21
0.10
0.33
0.49
0.37
0.41
0.29
Notes: Summary statistics are presented by migration status on the sample of working men in the private sector.
Migrants and incumbents account for 18,350 and 165,967 observations, respectively.
35